LVIO-GSMAP: Robust LiDAR-Visual-Inertial UAV System With Gaussian Mapping

This paper presents a robust SLAM framework that integrates LiDAR-Visual-Inertial Odometry (LVIO) with 3D Gaussian Splatting (3DGS) to address critical challenges in outdoor environments, including photometric degradation, weak textures, and motion blur. Traditional LiDAR-Visual-Inertial Odometry systems struggle under these conditions, while existing 3DGS-based SLAM methods are limited to indoor scenes due to their reliance on dense RGB-D inputs. Our approach introduces two key innovations: First, a state estimation pipeline that transitions entirely to LiDAR-Inertial Odometry (LIO) when photometric reliability drops, leveraging LIVO architecture to achieve millimeter-level accuracy and robustness. Second, an outdoor-capable 3DGS reconstruction system that utilizes geometrically consistent LiDAR point clouds from the LVIO frontend, overcoming the scalability limitations of neural implicit representations in large-scale environments. Our system achieves real-time performance at 25 Hz while maintaining high-fidelity mapping. This work bridges the gap between robust odometry and photorealistic reconstruction for outdoor SLAM, enabling reliable operation in complex, outdoors environments.